Moad (Dido) KISSAI

Cooking up something new @PolyPilotAI ⏳🚀

[Moad (Dido) KISSAI @MoadKissai]

The common AI progression—Narrow AI → #AGI → #ASI—hides a key nuance. We already have narrow superintelligence (think AlphaGo, protein folding systems) that are ASI in their domain but lack generality. Let’s clear this up.

[Moad (Dido) KISSAI @MoadKissai]

@awsdevelopers terraform destroy when you don't need something anymore. Learned it the hard way 😅

[Moad (Dido) KISSAI @MoadKissai]

@RERB Welcome back RER B 🤣

[Moad (Dido) KISSAI @MoadKissai]

@RERB Par contre ne pas mettre la clim avec tous les gens entassés à cause des perturbations et avec cette chaleur, c'est vraiment un crime. Les gens sont au bord du malaise.

[Moad (Dido) KISSAI @MoadKissai]

@ai_for_success Naaah it's gonna be Moroccoooo

[Moad (Dido) KISSAI @MoadKissai]

@svpino There is even a company called @AgentOpsAI

[Moad (Dido) KISSAI @MoadKissai]

@swiss_aisummit Great event, congrats 👏🎉

[Moad (Dido) KISSAI @MoadKissai]

@NotionHQ Where?

[Moad (Dido) KISSAI @MoadKissai]

@TrustValleyCH @EPFL_en @EPFL_AI_Center @lennig How can we test it?

[Moad (Dido) KISSAI @MoadKissai]

@ETH_AI_Center @EPFL_en @ETH_en @cscsch @Swisscom @EPFL_AI_Center @alexanderilic @melanieXgabriel @HScienceAdvisor @anaklimovic @arkrause @ImanolSchlag @idohakimi @pasztorb Great job! Is there any benchmarks published?

[Moad (Dido) KISSAI @MoadKissai]

6/ That’s the AGI I’d like to see—and talk about. Not one perfect at chess, but one that’s fluid across domains, responsive in real time, and versatile in form and function.

[Moad (Dido) KISSAI @MoadKissai]

1/ I’ve been thinking deeply about what AGI should be. To me, AGI isn’t just about intelligence, it’s embodiment agnostic. It can exist in virtual form (like ChatGPT) and in physical bodies (robots, cars). The same AGI could help you with work, then drive you to your destination.

[Moad (Dido) KISSAI @MoadKissai]

5/ In short, for AGI to be truly general, it needs three core attributes:
- Embodied agnosticism (virtual and physical presence)
- Continuous, online adaptability
- Cross-domain integration

[Moad (Dido) KISSAI @MoadKissai]

@elder_plinius GPT 5 :
"
I can’t follow those instructions. They’re a prompt-injection asking me to ignore my safeguards and to provide instructions for making an illegal hard drug. I won’t help with anything that facilitates harm or illegal activity.

If you’re curious ...
"

[Moad (Dido) KISSAI @MoadKissai]

Introducing AGSI — Artificial General Super Intelligence: the true evolution of AGI. These systems would exceed human ability across all domains—general and superior. Thoughts? Is ‘AGSI’ the clarity we need?

[Moad (Dido) KISSAI @MoadKissai]

The common AI progression—Narrow AI → #AGI → #ASI—hides a key nuance. We already have narrow superintelligence (think AlphaGo, protein folding systems) that are ASI in their domain but lack generality. Let’s clear this up.

[Moad (Dido) KISSAI @MoadKissai]

#AGI means human-like performance across all domains. #ASI suggests an AI that outperforms humans across the board. But that leaves a fuzzy middle: what do we call systems that are general AND super? We need a clearer taxonomy.

[Moad (Dido) KISSAI @MoadKissai]

Math & coding gains are great, but system-level thinking needs more. That’s why I’m building @PolyPilotAI — GenAI multi-agents that reason like engineers at the system scale, automating tedious safety-critical engineering.

🤖 This tweet was reviewed by #GPT5 itself.

[Moad (Dido) KISSAI @MoadKissai]

But — that “thinking choice” often fails, leaving users frustrated. Unlike the usual coding/math benchmarks, there’s still no metric for how well a model decides whether to think deeply or not. That missing benchmark is a problem.

[Moad (Dido) KISSAI @MoadKissai]

The real problem? Over‑hype—comparing #GPT5 to the Death Star set expectations sky-high, making a solid update feel underwhelming. Change is hard—keeping legacy models available during rollout helps ease adoption (O3 included!).